Skip to content

Determining cause of error?

4 messages · rkevinburton at charter.net, Charles C. Berry, Henrik Bengtsson

Original
rkevinburton at charter.net
# 
This is both a specific question and a general one. First, I am running 'fitdistr' from library(MASS) and I get the following:

Error in fitdistr(templist, "weibull") : optimization failed

What is the cause of the error? How can I tell? Can I just catch this error, report it and move to the next set of data (eat the exception)?

Thank you.

Kevin
Charles C. Berry
#
On Wed, 13 Aug 2008, rkevinburton at charter.net wrote:

            

      
      
      
    

        
You can use try() or tryCatch() to both allow further calculation and to 
save the current input to fitdistr() for further study.

Alternatively, you can set options(error=recover) - see ?recover - 
and see where the error comes from.

HTH,

Chuck

            

      
      
      
    

        

      
      
    

        
Charles C. Berry                            (858) 534-2098
                                             Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu	            UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901

  
  


  


      

    

    

      
      

        


  

        

      Henrik Bengtsson
    

    

      
      #
    

  


  

      
See ?traceback

/Henrik

        
On Wed, Aug 13, 2008 at 7:33 PM,  <rkevinburton at charter.net> wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      rkevinburton at charter.net
    

    

      
      #
    

  


  

      
Now this is really specific. I think the cause of the error is a small sample size. For example. The following both fail:

fit <- fitdistr(c(120), "weibull")
fit <- fitdistr(jiitter(c(120,120), amount=0.5), "weibull")

As it is hard for me to control the sample size or the proximity of data values in the sample I was wondering how I can tell what the error modes for 'fitdistr' are? If I knew that 'fitdistr' will error out with a sample of of one or a sample of two values that are close then I could put a conditional statement an make a "reasonable" guess (something that I would expect 'fitdistr' to do). For example I could"

if(length(templist) > 1)
{
    fit <- fitdistr(templist, "weibull")
}

The problem is that I don't know all the possible error conditions. I know that having a sample size of 1 is bad and a sample of two "close" values is bad. Are there others?

Thank you.

Kevin

        
---- Henrik Bengtsson <hb at stat.berkeley.edu> wrote: